KR100752066B1 - Method and device for coding/decoding - Google Patents

Abstract

The present invention relates to a method and a coder / decoder for coding / decoding a target range from a value range, in which the target range to be decoded until or until the code bit for the target range to be coded is found. Repeat the cyclic step in the coding / decoding step for the target range until. In the case of a cyclic step, the interval of the value range in which the target range to be coded / decoded is located is divided into two new intervals, and a single bit signals where the target range to be coded / decoded is located among the new intervals. The new interval thus signaled is used as the interval for the next cyclic step. The code bits for the target range to be coded or the target range to be decoded are found when the interval falls below the minimum size. At least one cyclic step is based on a probability distribution for the target range within the interval, and new intervals are selected based on the probability distribution.

Value range, target range, code bits, cyclic steps, signaling, probability distribution

Description

Coding / Decoding Method and Device {Method and device for coding / decoding}

The present invention relates to a coding / decoding method and apparatus using an iterative cyclic step.

For each cyclic step, the interval at which the value to be coded is divided into two new intervals, and which of the new intervals formed is represented by the bit. The method continues to run until the value to be coded / decoded is located.

The method and apparatus according to the invention having the features of the independent claims have the advantage of keeping the number of code bits small.

Other advantages derive from the measures set out in the dependent claims. In particular, the method is preferred when coding / decoding a plurality of data having a specific sequence. In addition, when the interval limit is properly selected, the required number of code bits can be significantly reduced. In particular, as a simple model, a probability distribution is used, which depends on how many elements follow in the sequence. The probability distribution thus measured has proven to be useful in certain applications. This is particularly true of audio data. The measured probability distribution can be stored particularly simply when the distribution is converted to an interval table.

An embodiment of the present invention is shown in the drawings and described in detail below.

1 shows a transmitter and a receiver.

2 illustrates selection of values by interval partitioning.

3 to 5 illustrate various probability distributions.

1 shows a data transmission system with a transmitting station 1 and a receiving station 2, which transmits data to the receiving station 2 via a transmission channel 3. For transmission, the data to be transmitted must be coded. That is, the transmitting station 1 converts the data to be transmitted into digital bit signals, which are then transmitted over the transmission channel 3. At the receiving station 2, the transmitted bit sequence is decoded as described above. That is, the original signal is formed again from the received bit sequence. Thus, in the following method, it is explained how the coding of specific data is performed particularly effectively, i.e., how much data amount is possible with a particularly small number of coding bits.

The data to be coded is illustrated in FIG. 2 by a stream of numbers representing a value range of 0 to 1023 (= 2 ¹⁰ −1). The information or data to be coded consists of selecting individual numbers from the value range of 0 to 1023 shown in FIG. This is illustrated by the number 330 in FIG. 2. The information, i.e. the individual data transmitted from the transmitting station 1 to the receiving station 2, consists of the number 330 being selected from the amount of numbers from 0 to 1023.

In addition, in FIG. 2, which number 330 is located by interval partitioning is described. In the first step, hereinafter referred to as the cyclic step, the maximum possible interval (first step = interval 0 to 1023) is divided into two new partial intervals. As a starting interval, the entire value range, i.e., numbers between 0 and 1023, is divided into two partial intervals by designating an interval limit of 512, where a first interval of 0 to 511 and a second interval of 512 to 1023 are obtained. By a single bit, it is indicated at which interval the coded number 330 is located, for example 0 can always indicate that the number to be coded is located in the lower interval, and 1 can indicate that the number to be coded is located in the upper partial interval. In the present embodiment, the first code bit may be 0 since the number 330 to be coded is located at a lower interval, that is, an interval of 0 to 511. In the second step, the interval defined above is subsequently divided into partial intervals. This is done by interval limit 256, one single bit signaling whether the number 330 to be coded is located between the lower interval, i.e., 0 to 255 or the upper interval, i.e., between 256 and 511. The code bit in this case is 1 since the number 330 to be coded is located in the upper interval. This interval is further divided into two new intervals, i.e. the lower interval of 256 to 384 and the upper interval of 384 to 512, in a further step. Code bit 0 indicates that number 330 to be coded is located at the lower interval. The method continues until the remaining interval width includes one number, that is, the number 330.

Decoding is done in the same way, until the receiving station 2 likewise partitions the interval by the received sequence of bits until the coded number 330 is found at intervals from 0 to 1023.

In the above method of dividing each observed interval into two new intervals of the same size, it can be readily seen that 10 bits are always needed to select or code one number from an interval of 0 to 1023. Thus, unlike using a dual system directly to transfer numbers, bits are not saved. However, this does not always give an interval limit by bisection of the interval as in this case, but changes when another interval limit for interval segmentation is selected based on the probability distribution. For example, if the number of codes to be coded is closer to zero than 1023, then the number of bits required can be reduced by properly selecting the interval limit at interval partitioning. For example, if the numbers to be transmitted are located at intervals of 0 to 63 with a probability of 0.5 and at intervals of 64 to 1023 with a probability of 0.5 remaining, the lower interval range of 0 to 63 and the upper interval range of 64 to 1024 are the first. Since it is sufficient for one-interval partitioning, the coding requirements can be reduced when statistically averaged. If the first code bit is zero, it can be clearly seen that the value to be coded is located at an interval of 0 to 63, and that the value to be transmitted can be coded with an additional six bits. By using the probability distribution for the values to be coded or decoded at interval partitioning, the bit cost required for coding can be reduced.

Another advantage is gained when transferring sets of multiple numbers. When transmitting a plurality of numbers from intervals 0 to 1023, i.e., when decoding correspondingly again after coding, the number of code bits required can be reduced by classifying the data. In this case, it is important that the numbers are transmitted, i.e. coded and decoded, in a sequence in which the numbers form a descending or ascending order. After coding of number 330, if it is apparent that the next number is larger or smaller, the interval required for interval partitioning can be reduced. The following description is based on the fact that the number 450 has to be transmitted in addition and the transmitter (coder) and receiver (decoder 2: decoder) recognize that two numbers are transmitted and that the first number to be transmitted is smaller. . Only an interval of 330 to 1023 needs to be observed in coding and decoding. Since this interval is less than an interval of 0 to 1023, fewer bits may be used in some cases. Thus, the number of bits required for coding can be reduced when a plurality of numbers to be transmitted (coded / decoded) are transmitted in sequence and the transmitter and receiver recognize this. For example, if the smaller number is 960, then only 64 numbers remain in the remaining intervals, which are simply coded into 6 bits. Since a plurality of numbers can be transmitted in a specific sequence, bits can be saved. It is also proposed that under this condition, the interval limit for interval partitioning does not have to consist of bisection of the remaining intervals.

3, 4 and 5 illustrate various probability distributions from theoretical considerations. Indeed, this probability distribution proved to be very good approximation for real data, especially real audio data to be coded.

3, 4 and 5, the probability distribution P (x) is shown in comparison with the normal interval x ranging from 0 to 1. For specific purposes, each of these normal intervals must be converted to a specific number range in which the numbers to be coded / decoded are located. In Figures 3 to 5, the distribution of numbers within one interval depends entirely on how many are within the interval. This is especially important when multiple numbers are coded / decoded. For example, if ten numbers from one value range are to be coded / decoded, then a different probability function is the basis than when only one number should be coded / decoded. In coding / decoding the numbers successively, this means that a different probability distribution should be used for each coding / decoding step. For this purpose, in coding and decoding it is necessary to know the total number of numbers to be coded or decoded. Then, under the assumption that the numbers are equally distributed in the entire value range, the probability distributions shown in Figs.

3 shows the probability distribution P (x) when the single value is coded / decoded. As can be readily seen, the probability distribution is constant over the normalized full value range of zero to one. In this case, the preferred interval limit is obtained by dividing the center. In Figure 3, the meaningful interval limits are shown by vertical lines, each given a percentage. 25% line when x = 0.25, 50% line when x = 0.5 and 75% line when x = 0.75. As can be easily understood, the 50% limit results in lower and upper intervals, in which case the probability that the value to be coded / decoded is in the lower interval is 50%, and the probability in the upper interval is 50% as well. In the first splitting step, if it is detected that the value to be coded / decoded is located in the lower interval, then the 25% limit, i.e., the limit of dividing the new interval into two intervals with 50% probability each, will define the next useful splitting limit. it means. FIG. 3 also shows that under the same distribution, little bit is reduced in coding / decoding individual values from one interval, since it is exactly the same as the partitioning method described in FIG.

4 shows the probability distribution for numbers at normal intervals of 0 to 1, assuming that additional numbers are coded / decoded and two numbers of independent equal probabilities are estimated over the entire value interval. The number to be coded is coded in a predetermined order, for example, first the small number is coded and then the large number is coded. To code the first number, the probabilities do not have the same magnitude over the entire value range of 0-1023. If the small number takes 1022, the large number should take the value 1023, but this does not mean anything except that the probability of the small number taking the value 1022 is relatively low. On the contrary, if the small number has one, the second number, that is, the large number has a value range of 1 to 1023. That is, the probability that the small number will take the value 1 is much greater than the probability that the small number will take the value 1022. The smallest, smallest probability curve at 1023 and the end of the value range, that is, at 1023, is correspondingly obtained. In the normalized value range of 0 to 1, the probability P (x) is largest at zero and smallest at one. Therefore, in order to code / decode two numbers using interval partitioning, the probability distribution according to FIG. 4 should be used for a small number and the probability distribution according to FIG. 3 for a large number. The distribution according to FIG. 4 is converted into a full value range by multiplication with a value range and the distribution according to FIG. 3 is converted by multiplication and a shift from a small value to the remaining interval extending from the upper limit of the value range. The interval limit should always be chosen so that the probability that one number is in two new intervals is equal. In this case, for this division, the first interval limit is shown as in FIG. 3. The 25% line is at x = 0.134, the 50% line is at x = 0.293 and the 75% line is at x = 0.500. Depending on at what interval the number to be coded / decoded is located, 50% lines are selected in the first division step and 25% lines or 75% lines in the second division step.

FIG. 4 shows the probability distribution at normal intervals of 0 to 1 for the number to be coded / decoded, assuming there are more numbers. In addition, assume that the numbers are coded / decoded in a sequence. Under this assumption, the limit is selected as indicated again in percent in FIG. 4 to determine the interval limit of interval partitioning. In practice, this interval limit is designed to be asymmetrical so that the interval is not bisected and the probability for being coded / decoded is 50% on the right and left sides of the interval limit, respectively.

FIG. 5 is still followed by the coded / decoded number, followed by two additional coded / decoded numbers, the value of which is greater than the number for the probability distribution of FIG. 5, for one number from the normal interval. , Probability distribution at the normal interval of 0 to 1 is shown. Under the above assumptions about the number to be coded / decoded, a probability distribution occurs where a small value (i.e., a value close to zero) is more likely than a value close to one. Based on this probability distribution, an interval limit represented by 25%, 50%, 75% can be selected here, so that a number requiring only a few bits can be coded very effectively. The 25% line is at x = 0.0914, the 50% line is at x = 0.206 and the 75% line is at x = 0.370.

Assuming that four values follow or five values follow, a corresponding probability distribution can be formed. However, as the number of values increases, the probability distribution changes only marginally, that is, the probability distribution changes only marginally starting from the limit of about 10 values.

If three values must be transmitted from one value range, these values are sorted in ascending order according to their size. That is, in the value range, the first number is the smallest value and the last number is the largest number. Under this assumption, the probability distributions of Figures 3, 4 and 5 can be used. To code / decode the first number, the probability distribution of FIG. 5 is used over the full value range. After this first number is coded / decoded, the probability distribution of FIG. 4 is used for coding / decoding of a second number greater than the first number. However, only the interval between the smallest number and the upper limit of the value range is considered. Thus, the probability distribution of FIG. 4 is normalized for this interval. Through interval partitioning considering the probability distribution of FIG. 4, a second number is coded / decoded within the interval. To code the largest third number, a probability distribution according to FIG. 3 is used in the interval, the lower limit of the interval being formed by the decoded second number, and the upper limit being formed by the upper limit of the value range. To code and decode this number, a probability distribution according to FIG. 3 is used, which in turn means always bisects the remaining intervals. With this measure, the number of bits needed to code three numbers is kept small.

Since the method is based on a probability distribution, the number of bits required for coding depends on how the actual distribution of the number matches the probability distribution. Thus, it is not always clear how many bits are needed to code a specific number of sets. If the number has a very low probability distribution in the value range, in individual cases, more bits may be needed than simple interval partitioning by bisection of the intervals in the present invention. However, statistically averaged, significant savings are achieved with bit rates reduced by 20% to 30%.

4 and 5 apply to numbers to be coded / decoded, followed by larger numbers to be coded / decoded. That is, it applies to ascending classification of the numbers to be coded / decoded. If a descending classification of the numbers to be coded / decoded is used, then the probability distribution is reflected accordingly. In other words, the probability distributions are the largest at value 1 and the smallest at value 0, respectively.

The probability distributions of FIGS. 3-5 are based on theoretical considerations regarding the distribution of values within one value interval. However, the actual values are shown to show similar dependencies. Here, the transmission of audio data is referred to as an example, and the transmission of the data is achieved by the frequency of the sinusoidal vibration being transmitted as a numerical value. The evaluation of the actual audio data indicates that after frequencies have been sorted in ascending or descending order, the probability of a particular frequency in the full range of possible values of frequency depends on how many additional frequencies are continuous in a row. This probability distribution shows a shape similar to that of FIGS. 5 and 6 discussed here. That is, depending on the number of additional frequencies that follow when classifying frequencies as a column of ascending numbers, it is shown that the probability toward smaller frequency values is greater than the probability toward larger frequency values. Sample probabilities are calculated by evaluation of the actual audio data, which has proved to be desirable for coding an ascending order of frequency numbers, each representing superimposed sinusoidal frequencies in order to reconstruct the original signal.

In order to use the probability distribution thus measured, it is not necessary to store the entire probability distribution, but rather it is sufficient to indicate that a meaningful interval limit for interval partitioning is based on a reference interval. The interval limit for the first five interval partitions, obtained from the selected audio data, is presented below. This is 31 consecutive values for an interval limit that splits the value range into 32 ranges of equal probability. The values mentioned are based on a value range of 1024. That is, for use in one specific value range, these numbers are divided by 1024 and then multiplied by the number of values in the interval considered in the specific use, respectively. The partitioning limit for dividing the probability distribution into 32 ranges of equal probability is as follows:

If there is no subsequent value,
53, 87, 118, 150, 181, 212, 243, 275, 306, 337, 368, 399

431, 462, 493, 524, 555, 587, 618, 649, 680, 711, 743, 774

805, 836, 867, 899, 930, 961, 992

If there is one successor

 34, 53, 71, 89, 106, 123, 141, 159, 177, 195, 214, 234

254, 274, 296, 317, 340, 363, 387, 412, 438, 465, 494, 524                 

556, 591, 629, 670, 718, 774, 847

If there are two subsequent values,

 26, 41, 54, 66, 78, 91, 103, 116, 128, 142, 155, 169

184, 199, 214, 231, 247, 265, 284, 303, 324, 346, 369, 394

422, 452, 485, 524, 570, 627, 709

If there are three successors

 23, 35, 45, 55, 65, 75, 85, 96, 106, 117, 128, 139

151, 164, 177, 190, 204, 219, 235, 252, 270, 290, 311, 334

360, 389, 422, 461, 508, 571, 665

If there are four subsequent values,

 20, 30, 39, 48, 56, 64, 73, 81, 90, 99, 108, 118

127, 138, 149, 160, 172, 185, 198, 213, 228, 245, 263, 284

306, 332, 362, 398, 444, 507, 608

If there are five subsequent values,

 18, 27, 35, 43, 50, 57, 65, 72, 79, 87, 95, 104

112, 121, 131, 141, 151, 162, 174, 187, 201, 216, 233, 251

272, 296, 324, 357, 401, 460, 558

If there are six subsequent values,

 16, 24, 31, 38, 45, 51, 57, 64, 70, 77, 84, 91

 99, 107, 115, 123, 132, 142, 152, 163, 175, 188, 203, 219                 

237, 257, 282, 311, 349, 403, 493

If there are seven subsequent values,

 12, 19, 25, 30, 35, 41, 46, 51, 56, 62, 67, 73

 79, 85, 92, 99, 106, 114, 122, 132, 142, 153, 165, 179

195, 213, 236, 264, 301, 355, 452

As described above, the probability distribution for a very large number does not differ from the subsequent value, so in these cases the probability distribution for seven subsequent values can always be used (which is at least valid in audio data). In addition, in the audio data, when the first five interval partitioning is performed in consideration of the probability distribution, additional savings are lowered when the probability distribution is further considered. Therefore, in the audio data, only the first five interval partitions are made under consideration of probability distribution, and in some cases, the remaining intervals are divided into two new partial intervals of the same size, so that only additional interval partitioning remaining is performed. Appeared to be sufficient.

Under the assumption that an additional number follows, the use of the table is described based on the coding of the number 330.

Wherein the number 330 is selected from an interval of 0 to 1023. Looking at the table, the number 317 is set as the first partition limit. This means that the probability that the number to be coded is below the limit is exactly the same as the probability that it is above the limit. Since the relative number 330 is above this limit, the first code bit is one. For the remaining intervals from 318 to 1023, the next split is made at 524, as shown in the table. Thus, the second code bit is zero. The next division is made at 412 (code bit = 0), ie the coded value is located at an interval of 318 to 412. The next division is made at 363 (code bit 0), ie the number that can be coded is limited to an interval of 318 to 363. The next partition (340 in the table) limits the range of numbers that can be coded to range from 318 to 340. Subsequent coding is simply done by dividing the remaining intervals. In addition, the division of the intervals 318 to 340 is made at the limit 329 = (318 + 340) / 2, so that the code bit is one. As a result, the remaining intervals of 330 to 340 are divided again centrally in subsequent cyclic steps, until the remaining intervals generated contain only one number to be coded (here 330).

The table stored as above allows, by very simple means, to reduce the bits required for coding. The amount of memory required for this is reduced by storing only the individual limits for the intervals, and the coding station 1 as well as the decoding station 2 need not make a great effort to use the method.

The method is described by a number selected from a range of values. Instead of numbers, any type of data placed may be used, which may be placed in sequence. Since such data is converted into corresponding numbers within one value range without problem, the method is suitable for any type of data. Also, a range can be used instead of individual numbers, that is, the number range can be simply coded.

The method is always executed until an individual number is reliably detected. The method may also abort when the remaining interval is small and can be accepted in each application. This can further reduce the bits required for coding, so that the bit rate can be varied in a very simple manner.

Claims (11)

 CLAIMS 1. A method for coding / decoding a target range from a value range, the method comprising: until the code bits for the target range to be coded are found or using the code bits until the target range to be decoded is found. The cyclic step is repeated in the coding / decoding step for the range, the interval of the value range in which the target range to be coded / decoded is located is divided into two new intervals in the cyclic step, and a single bit is the target to be coded / decoded. A range is used to signal the new interval at which the range is located, and this signaled new interval is used as the interval for the next cyclical step, and when the interval has not reached the minimum size, the target range to be coded or the target range to be decoded. The code bits for the at least one cyclic step are found From, based on the probability distribution for the target range, the new interval will coding / decoding method, characterized in that is selected based on the probability distribution. A coding / decoding method according to claim 1, wherein an individual number is used as said target range. 3. A method according to claim 1 or 2, wherein in each cyclic step the interval is selected such that the target range to be coded / decoded is located in each of the two new intervals with equal probability. 3. A method according to claim 1 or 2, wherein a plurality of said target ranges are coded / decoded, the number of said target ranges is given as a number of values, and wherein said coding / decoding step is characterized in that all said target ranges of number of values are coded / decoded. Coding / decoding method which is repeated until decoded. 5. The method of claim 4, wherein to detect the probability distribution, a sequence of the target ranges to be coded / decoded is determined, and the probability distribution of one target range depends on how many target ranges follow in the sequence. A coding / decoding method characterized by the above. 6. The method of claim 5, wherein the target range to be coded / decoded forms an ascending or descending order. 6. A method according to claim 5, wherein a plurality of probability distributions are stored and the selection of the probability distribution for one coding / decoding step depends on how many of the target ranges follow in the sequence. . 8. The method of claim 7, wherein the stored probability distribution is detected from a measurement of data. 8. The coding / decoding method of claim 7, wherein the stored probability distribution is calculated. 3. Method according to claim 1 or 2, characterized in that the number of coded / decoded is an integer.  A coder / decoder for coding / decoding a target range from a value range, wherein the target is either until a code bit for the target range to be coded is found or the code bit is used to find the target range to be decoded. Means for repeating a cyclic step in a coding / decoding step for a range; Means for dividing an interval of a value range in which the target range to be coded / decoded is located into two new intervals in the cyclic step; Means for indicating using a single bit where the target range to be coded / decoded is located among the new intervals; Means for using the new interval indicated by the single bit as an interval for a next cyclic step; And setting that the code bit for the target range to be coded or the target range to be decoded has been found comprises means for determining when the interval falls below a minimum size; And means for selecting two new intervals in at least one cyclic step using a probability distribution over the target range within the interval.

Images